Learning

Learning is handled by various classes :

• The Learner is the main class that runs the full process

• Listeners are used for validation or other monitoring tasks

• Trainers that iterate over batches of data

• Optimization deals with parameters (selecting, gradient descent, etc.)

Learner

The main class is the Learner task; when submitted to the scheduler, returns a LearnerOutput.

XPM Taskxpmir.learning.learner.Learner(*, random, trainer, model, max_epochs, steps_per_epoch, use_fp16, optimizers, listeners, checkpoint_interval, device, hooks, use_pretasks)

Bases: Task, EasyLogger

Submit type: xpmir.learning.learner.LearnerOutput

Model Learner

The learner task is generic, and takes two main arguments: (1) the model defines the model (e.g. DRMM), and (2) the trainer defines how the model should be trained (e.g. pointwise, pairwise, etc.)

When submitted, it returns a dictionary based on the listeners

random: xpmir.learning.base.Random

The random generator

trainer: xpmir.learning.trainers.Trainer

Specifies how to train the model

model: xpmir.learning.optim.Module

Defines the model to be learned. If multiple models are used, one can use MultipleModel.

max_epochs: int = 1000

Maximum number of epochs

steps_per_epoch: int = 128

Number of steps for one epoch (after each epoch results are logged)

use_fp16: bool = False

Use mixed precision when training

optimizers: List[xpmir.learning.optim.ParameterOptimizer]

The list of parameter optimizers

listeners: List[xpmir.learning.learner.LearnerListener]

Listeners are in charge of handling the validation of the model, and saving the relevant checkpoints

checkpoint_interval: int = 1

Number of epochs between each checkpoint

logpath: Pathgenerated

The path to the tensorboard logs

checkpointspath: Pathgenerated

The path to the checkpoints

device: xpmir.learning.devices.Device = xpmir.learning.devices.Device.XPMValue()

The device(s) to be used for the model

hooks: List[xpmir.context.Hook] = []

Global learning hooks Initialization hooks are called before and after the initialization of the trainer and listeners.

use_pretasks: bool = False

Use deprected pre-tasks as the output

namedtuple xpmir.learning.learner.LearnerOutput(listeners: Dict[str, Any], learned_model: ModuleLoader)

Bases: NamedTuple

The data structure for the output of a learner. It contains a dictionary

where the key is the name of the listener and the value is the output of that listener

Fields:

0.  listeners (Dict[str, Any]) – Alias for field number 0

1.  learned_model (ModuleLoader) – Alias for field number 1

Trainers

Trainers are responsible for defining the the way to train a learnable scorer.

XPM Configxpmir.learning.trainers.Trainer(*, hooks, model)

Bases: Config, EasyLogger

Submit type: xpmir.learning.trainers.Trainer

Generic trainer

hooks: List[xpmir.learning.context.TrainingHook] = []

Hooks for this trainer: this includes the losses, but can be adapted for other uses The specific list of hooks depends on the specific trainer

model: xpmir.learning.optim.Module

If the model to optimize is different from the model passsed to Learn, this parameter can be used – initialization is still expected to be done at the learner level

XPM Configxpmir.learning.trainers.multiple.MultipleTrainer(*, hooks, model, trainers)

Bases: Trainer

Submit type: xpmir.learning.trainers.multiple.MultipleTrainer

This trainer can be used to combine various trainers

hooks: List[xpmir.learning.context.TrainingHook] = []

Hooks for this trainer: this includes the losses, but can be adapted for other uses The specific list of hooks depends on the specific trainer

model: xpmir.learning.optim.Module

If the model to optimize is different from the model passsed to Learn, this parameter can be used – initialization is still expected to be done at the learner level

trainers: Dict[str, xpmir.learning.trainers.Trainer]

The trainers

XPM Configxpmir.letor.trainers.LossTrainer(*, hooks, model, batcher, sampler, batch_size)

Bases: Trainer

Submit type: xpmir.letor.trainers.LossTrainer

Trainer based on a loss function

This trainer supposes that:

• the sampler_iter is initialized – and is a serializable iterator over batches

hooks: List[xpmir.learning.context.TrainingHook] = []

Hooks for this trainer: this includes the losses, but can be adapted for other uses The specific list of hooks depends on the specific trainer

model: xpmir.learning.optim.Module

If the model to optimize is different from the model passsed to Learn, this parameter can be used – initialization is still expected to be done at the learner level

batcher: xpmir.learning.batchers.Batcher = xpmir.learning.batchers.Batcher.XPMValue()

How to batch samples together

sampler: xpmir.learning.base.Sampler

The sampler to use

batch_size: int = 16

Number of samples per batch

process_microbatch(records: BaseRecords)

Combines a forward and backard

This method can be implemented by specific trainers that use the gradient. In that case the regularizer losses should be taken into account with self.add_losses.

Listeners

Listeners can be used to monitor the learning process

XPM Configxpmir.learning.learner.LearnerListener(*, id)

Bases: Config

Submit type: Any

Hook for learner

Performs some operations after a learning epoch

id: str

Unique ID to identify the listener (ignored for signature)

__call__(state: TrainState) → LearnerListenerStatus

Process and returns whether the training process should stop

XPM Configxpmir.learning.context.ValidationHook

Bases: Hook

Submit type: xpmir.learning.context.ValidationHook

Base class for all the validation hooks